AI execution architecture

Deterministic programs first, single agents first; only complex tasks add specialist roles and multi-model collaboration. Permissions are checked before a call and evidence after the run.

In developmentNot availableVerified

AI execution architecture diagram (A05) (open full-size figure)
MYRILUM architecture map v2.2 · A05 (public edition). Labels are in Chinese; every element is listed in English below. Select the diagram to open it at full size.

What this map shows

This is MYRILUM's target architecture. It does not mean everything on the map is live; whether a capability can be used today is stated in “Current availability”.

Deterministic programs first, single agents first; only complex tasks add specialist roles and multi-model collaboration.

Permissions are checked before a capability is called and evidence after the run ends; a model can claim neither for itself.

Execution inputs and boundaries

Work orderInput references, output standards, task version and dependencies
Scoped authorizationTool allowlist, data permissions, recipients and validity period
Budget and timeReserved credits, call limits and deadline
Check contractArtifact format, evidence requirements, and failure and stop conditions

Processes and run instances

WorkflowDeterministic steps, branches, waits and recovery
Agent definitionRole, skills, model policy and version
Agent runtimeState, context, leases, attempts and checkpoints
Professional human taskA clear task sheet, deadline, handoff and acceptance

Capabilities through unified gateways

Model gateway and routingWeighs quality, cost and latency among permitted candidates
Tool gatewayParameter contracts, authorization, idempotency and side-effect control
Controlled knowledge and memoryScoped at recall; permissions are checked again before returning
Remote execution adaptersAPI / MCP / A2A; a protocol never replaces the business contract

Results enter the business loop

Call and usage receiptsProvider request identifiers, versions, resource amounts and exceptions
Candidate resultsContent, data, files and source references
Verification and risk checksFailures go back for revision; major uncertainty goes to a person
Controlled state commitDomain rules commit; models never write official facts directly

Source

This page follows A05 of the MYRILUM architecture map v2.2 (public edition: restricted products and internal open items are left out).

Was this page helpful?

Sources and maintenance
Region
Global
Version
0.1.0
Publication
Published globally
Document status
Approved · Validated
Authorization
Public information
Command authority
None
Type
Concept
Audience
Technical leads, Developers, Organization leaders, Integration partners, Agents
Safety class
Informational
Agent tasks
understand_platform_architecture
Owner
Web3Capital Documentation Steward
Last verified
2026-09-23
Review due
2026-10-23
Source commit
882487dec08b
Canonical authority
SRC-ARCHITECTURE-ATLAS-V2-2
Content digest
9612de086c7746d1
Stable citation
MYRILUM-DOC-ARCH-005@0.1.0:en#overview

Machine-readableMarkdownMetadataJSON-LDChunks

This documentation grants no execution, release, or production authority.

Sources and maintenance

Region
Global
Version
0.1.0
Publication
Published globally
Document status
Approved · Validated
Authorization
Public information
Command authority
None
Type
Concept
Audience
Technical leads, Developers, Organization leaders, Integration partners, Agents
Safety class
Informational
Agent tasks
understand_platform_architecture
Owner
Web3Capital Documentation Steward
Last verified
2026-09-23
Review due
2026-10-23
Source commit
882487dec08b
Canonical authority
SRC-ARCHITECTURE-ATLAS-V2-2
Content digest
9612de086c7746d1
Stable citation
MYRILUM-DOC-ARCH-005@0.1.0:en#overview

This documentation grants no execution, release, or production authority.